### Multilayer Social Networks, chapter 5 (book)

This book chapoter originall appeared as this paper

Primary Reviewer: Guy

Secondary Reviewer: Fintan

#### Primary Review:

The authors somehow echo a taxonomy of visual approaches they classify into four main groups (chapter 5, page 80):

• those focusing on structure (pure node-link layout, for instance)
• those focusing on network statistics (node degree bar chart, for instance)
• augmented representation combining structure together with some attributes (revealed using a visual variable)
• simplified visualization where structure is revealed using some filtering/aggregation technique

It is important that we either agree on this taxonomy of use another one. This wold help us classify the techniques we collect and present as being relevant for multilayer networks.

##### Tasks typical of multilayer networks
• A main task that in a sense justifies why multilayer networks are worth considering is that of understanding how various layers relate. Many domain questions indeed turn out to be instances of this high-level question. Being able to define roles in a social network, being able to identify operational mode in criminal networks, etc.

The following list of tasks is inspired (or bluntly copied from) Chapter 5 of the book.

• Represent a same metric (network statistics) on different layers (section 5.2)
• Show variations of layer metrics for a same actor.
• Other variations are possible: show how different metrics/different layers compare
• Etc.
• Investigate the correlation between layers
• Analyze correlation between any two subsets of layers
• The previous tasks calls for being able to aggregate layers
• Providing feedback (quantitative, qualitative) on edge co-occurrence between layers contributes
• Visualize distribution of multiplex patterns
• Frequency of complex multiplex patterns involving numerous layers
• Keep track of layer-related information (through any visualization)

#### Secondary Review

This chapter is a summary over viewer of some technique for visualizing multilayer networks It is a survey and appears to refer to other people’s work, so it should not be part of our literatures review [Not sure I agree. On the contrary, I believe we must cite this source and explain in what manner it comes short of reporting on previous work.], but it is very interesting and is definitely worth while reading. They discuss four main approaches : Structure (e.g. NL diagram.), Metrics (degree, betweenness etc.) , augmented vis ( a mix of the first two) and simplified vis, e.g. showing k core network.This is and interested subdivision of vis types, but I think we will probably be mere details, and look at interaction etc.… The chapter discusses metrics (we describe it as attribute visualization with respect to cross layer visualization and goes into detail about parallel coordinates and variants thereof. Figure 5.2 a is a very interesting variant (which they do not cite the source for) for showing attribute scores across layers. They also discuss using matrices to visualize attributes. For showing the relationships between layers, they give examples of some interesting edge co-occurrence displays, which may be useful for layer comparison. The discussion on sliced layout may also provide some useful input for BLIZAAR, essentially it is a small multiples approach. It is worth nothing that the authors also examine Mux vis’ layer slice approach, and are not critical of it at all. Overall this is an interesting summary of multilayer vis, but it generally sticks to established approaches [Agree.]. The discussion on metric visualization is good. However, the chapter does not consider interaction at all. It does however discuss multi layered network vis in an interesting and relevant manner. I think that this chapter can be useful for everyone writing [reading?]our survey

### On open problems in biological network visualization

Primary Reviewer: Guy

Secondary Reviewer: Sebastien

#### Primary Review:

Biological networks provide a good application domain example – we kind'a knew it, didn't we? According to the authors, biological data is organized or seen as primary or secondary data. Layers consist in nodes and edges of different types. Examples are:

• Protein-protein or protein-DNA interaction
• Biochemical reactions or inter-pathway connections

### Visualization techniques for categorical analysis of social networks with multiple edge sets

Primary Reviewer: Fintan

Secondary Reviewer: Bruno

#### Primary Review:

This paper from the social network analytics field provides nothing groundbreaking in terms of visualization approaches, but does at least provide some interesting presentations from a sociologist POV, and interesting applications of existing ideas. The graph layers are defined by relationship type. The visualization features enhancements to basic radial layouts, using bundling and clustering and routing intra-group edges outside of the circle. The sensitivity analysis approach to regular graph layout is not multilayer specific. It is a paper worth mentioning as a social networks example and example of edge routing / bundling. There is no novel interaction or anything too exciting about layer definition

#### Secondary Review

This paper is about social network analysis and how to cope with the problem of large and dense graphs which look like a “hair-ball” when drawn with a standard force-layout method. Authors applied sensitivity analysis (aka centrality measures) to show only important elements of the graph, modularity clustering to draw a skeleton instead of the whole graph, edge bundling, radial layout and parallel coordinates to visualize the resulting graphs.

Multilayer aspects are the in the dataset used where nodes and edges are labelled with many variables. This paper is interesting because of the many methods used to analyse data and how standard method can be applied to multilayer graphs. Only the analysis techniques are tweaked to handle multilayer. I have never thought radial layout could be useful. for this kind of task.

### MuxViz: a tool for multilayer analysis and visualization of networks

Primary Reviewer: Fintan

#### Primary Review:

This paper describes a toolkit for visualizing multilayer networks, and is a collaboration with many of the authors of the Multilayer networks paper. The provide example using multiple data sets (primarily biological, but some geographic ones are mentioned too). The primary visualization offered by the toolkits are:

• 2.5d visualization of the layers as node link diagrams, as well as a standard 2D view of the whole network. Nodes and edges can be coloured to show values such as the node and edge type or community membership. This pseudo 3D like approach to visualization is generally rejected by the vis community
• An annular ring visualization of metrics. Concentric rings with segments coloured by value, matching angles across layers represent the same node. This is used to show the relative values of centralities across the value of the graph as a whole (each ring is a different centrality) or more frequently used to compare centralities across layers. Of course the outer layers are easier to read, thean the inner ones. I would say this type of visualization is quite perceptually flawed.
• Simple distance matrices, with dendograms that show the compressibility of layers ( how similar they are), as well as heat map matrices summararies to summarise relationships between layers, such as node overlap , associativity etc.

The most important thing about this paper is that the visualizations are targeted specifically at multilayer analytics. The toolkit has a strong analytics based focus, one of the primary purposes of the visualization is to compare analytics results (centralities etc.) across layers. They also use some interesting analytics to better understand similarity between layers. They use information theory to consider compressing layers ( to determine which ones can be merged without losing much data), as well as looking at cross layer summary metrics and visualising them using matrices Interaction is not discussed, (this demonstrates a gap in the research I think), as usual there is no empirical evaluation. The lack of empirical evaluation is a big problrm as both the 2.5 layouts and the annular visualizations have known perceptual issues.

— -

### A Visual Analytics Approach to Dynamic Social Networks

Primary Reviewer: Sebastien

Secondary Reviewer: Fintan

#### Secondary Review

The graphs are multilayer in this paper as the authors consider the problem graphs to be possibly multi relational (edges) and multimodal (nodes). The layout is a variant for general dynamic graph layout, which . In terms of visual presentation , many of the common approaches , which may be adaptable for multiple layer presentation are adapted (juxtaposition, superimposition, 2.5D, and combinations). For these presentations the “layers” are the time slices. In terms of analytics, colour is used for values , such as eigenvector, which is nothing ground breaking. However the author do appear to like the use of their 2.5 view for trajectories of nodes across time-slices, which may be useful for any discussion on 3D style views.The interaction is not especially novel , just 2 variants on node highlighting There is no evaluation, so this paper may be referenced as an example for some common techniques but not validation of it. While the authors do mention terms such as multi-relational and multimodal as part of their problem definition, they never really address them specifically.

This paper has over 50 citations

### Visual analysis of overlapping biological networks

PDF link ( Version published at IV is to big to upload, but there is no difference in the content)

Secondary Reviewer: Fintan

#### Primary Review:

This paper proposes a constrained graph layout method for heterogeneous overlapping networks, consisting in fixing the graph layouts in one or two layers drawn in parallel planes in a 2.5D fashion, and subsequently in applying rotations and scaling on the graph layout in an additional free layer in order to minimize the total inter-plane edge length. The usefulness of the technique is illustrated using biological networks (gene regulatory networks, Protein interaction networks and metabolic networks). The concept of overlapping network is based on the sharing of a subset of edges or nodes. Authors draw a distinction between dynamic networks where “a set of similar networks arise from one network” and overlapping networks “consisting of different types of networks”. The graph sizes considered are in one case in the range of 20-50 nodes and about the same number of links and in a second case 100-1000 nodes and a few thousand links.

#### Secondary Review

This paper focuses a lot on the biological aspects and on the creation of the 2.5d planes and the positioning of the nodes in each plane. It is a worthwhile example of early attempts at multilayer graph visualization, but not interesting for future work. The layers are related to biolocical fucntionaly and Nodes which exist in more than one layer are connected by “inter plane edges”.There are no novel interaction or analytics techniques, nor is there any evaluation performed.

### Manynets: An interface for multiple network analysis and visualization

Primary Reviewer: Bruno

Secondary Reviewer: Sebastien

#### Primary Review:

ManyNets is a tool which helps easily to subdivise a network (use cases done on social networks and ego networks) in subnetworks and see them inside an (much) enhanced spreadsheet-like GUI. Analysing a network follows the Schneiderman Mantra. Each line is a subnetwork and columns are used to display its properties (structural or user defined). The top of each column shows an aggregated graphical plot of the property for all graphs. The tool is well adapted to visually study groups of subnetwork for revealing temporal pattern or correlations.

For this paper, multi-layer resides in the original network subdivision and not in the data. Each subnetwork may be studied independently or in user-defined groups via many possible aggregations. Multi-layer (even if it is not written) is here the good idea for a dense network analysis tool where an overview of the whole network does not bring any useful information.

### Multimodal Networks: Structure and Operations

Primary Reviewer: Guy

Secondary Reviewer: Sebastien

#### Primary Review:

The paper presents multimodal network as a novel graph-theoretic formalism used to describe biological networks.

### NodeTrix: a hybrid visualization of social networks.

Primary Reviewer: Sebastien

### Exploring the design space of composite visualization

Primary Reviewer: Sebastien

### DiffAni: Visualizing Dynamic Graphs with a Hybrid of Difference Maps and Animation

Primary Reviewer: Sebastien

Secondary Reviewer: Bruno

### TreeMatrix: A Hybrid Visualization of Compound Graphs

Primary Reviewer: Sebastein

Secondary Reviewer: Fintan

### Visual Mining of Multi-Modal Social Networks at Different Abstraction Levels

Primary Reviewer: Fintan

Secondary Reviewer: Sebastien

#### Primary Review:

The authors present a tool called INVENIO which looks at multi-modal, multirelational, multi-featured graphs. The visualization is standard toolkit NL diagrams with no enhanced interaction. There are different nodes types but only a single view (or layer). Might be worth mentioning the induction of subgraphs based on attributes and multiple abstraction levels but there is nothing very interesting beyond that for our survey here, the work has not dated very well.

### Jigsaw: Supporting investigative analysis through interactive visualization

Primary Reviewer: Bruno

Secondary Reviewer: Fintan

#### Secondary Review

Jigsaw is a document exploration tool that was one of the earlier papers to popularise graph list views They describe heterogenity across documents and do use the tem layer once, But not really as a strong concept The layers are defined by entity type. The views in the application are List View ,Graph (NL) view, and a scatterplot view . List view was interesting for the time and uses a scented widget to convey appearance frequency. The graph NL view does not show all nodes. Node must be clicked to be expanded, to allow avoid overwhelming the user. The scatterplot view would more accurately be called a matrix view. All interaction techniques seem quite standard The primary task is to explore interactions between entities across documents. A good solid paper, with some fundamental ideas implemented well , but it is not hugely innovative by modern standards. However we should definitely include it in our survey as it does address some important ideas

### Visualizing the Evolution of Communities in Dynamic Graphs

Primary Reviewer: Sebastien

Secondary Reviewer: Bruno

#### Primary Review:

The technique is based on existing ones (Sankey diagram, code flow, …) and thus may show evolving social communities. This may be used to show, for instance, the splitting of a large communities into smaller ones, or people joining new communities. But one novelty is that it integrates node-link diagrams showing (small) network structures of social communities at some points in time, along with its evolution. This may help make sense of the data e.g. were people leaving a community very connected to the others, or isolated ? In this paper, community of persons may be considered a layer, and they can show links across layers (inter-layers). Use case include soccer matches over time (teams being connected if they played together), social network of people interactions. Use cases are limited to show that the tool work for some cases (no evaluation or tasks). The technique is inspiring but very general (focused on dynamic networks) and may be less scalable ; it may need to be adapted to grasp real-world multilayer (& dynamic) networks.

### Design Study of LineSets, a Novel Set Visualization Technique

Primary Reviewer: Guy

Secondary Reviewer: Fintan

#### Secondary Review

A set visualization like bubble sets or g-map…..can be used to highlight clusters or possibly layers…. However it just selects nodes and not edges. Essentially a line set is a bézier curve based large coloured line which intersects all elements of the set. Was part of the survey due to the faceted key word. However it may be useful as a basis of showing multiple layers in one diagram, particularly for problems related to layer membership and intersection of layers.

### Visual Analysis of Large Heterogeneous Social Networks by Semantic and Structural Abstraction

Secondary Reviewer: Fintan

#### Secondary Review

This paper describes an application called Ontovis. The key idea is that an ontology graph ( essentially a graph of node types) is use to explore a much larger data set The authors use an ontology graph consisting, of the node types from the data set, to build a standard undirected graph reflecting the structure of the data set. The two different graph types could be considered types of layer. Additionally the different node types could be considered layers. In this case the ontological graph can then be considered to be a visualization of the layers. The visualization is nothing special, a standardish NL layout with curved edges, however as interaction goes The addition of nodes to the main graph by selection of the corresponding types in the ontology is quite novel and relevant to the definition of layers in a multilayer graph. Essentially the ontology graph is a type of attribute visualization. An interesting feature is that when an ontology node is selected to be added to the structure graph, resulting in nodes being created for every different vale of that ontology node , all the other types of nodes become attributes of the ontology nodes. The primary tasks explored how do different entity types interact with each other, which again may be useful for ML Graphs.The paper uses some interesting metrics concerning dispersion of average connectivity per node type which might be interesting for investigating general multilayer graph relationships . The approach of building a graph (or a layer ) based on ontology, may well be useful for multilayer graph applications ( in the digital humanities for example) Although not as clearly multilayer vis as some other papers, it is very interesting, especially with respect to interaction,

### FacetAtlas: Multifaceted Visualization for Rich Text Corpora

@article{Nan2010,

 author = {Nan, Cao and Jimeng, Sun and Yu-Ru, Lin and Gotz, D. and Shixia, Liu and Huamin, Qu},
title = {FacetAtlas: Multifaceted Visualization for Rich Text Corpora},
journal = {IEEE Transactions on Visualization and Computer Graphics},
volume = {16},
number = {6},
pages = {1172-1181},
year = {2010}

}

Primary Reviewer: Fintan

Secondary Reviewer: Guy

#### Secondary Review

The paper presents an approach to navigate text corpora through clouds (graphs) of words. Words map to various level of generalities and act as shared concepts between lower level words. The paper introduces the notion of facets close to the notion of clusters of words. Entities (lowest level words) sit “within” facets (paragraph before section 3.2). That is, it seems words may belong to multiple facets (section 3.2, “we extract multifaceted entities”).

The paper mises a clear definition of what facets, and what layers are. The paper also uses the notion of a cluster but does not clarify how clusters relate with facets and layers. Images suggest what they use are hierarchical clusters. The authors indeed talk about primary and secondary facets as if they only consider cluster hierarchy of depth 2. Entities map the lowest level elements (nodes).

They consider multiple types of relationships between words, but I could not tell whether types correspond to edges connecting entities with a same facets or between different facets. That being said, it seems part of the distinction also comes from considering edges as being internal or external.

• Being able to “switch context” – again without being clear on what context really is.
• Pivot the primary visualization layout arrangement across different facets. (Although this is less a domain task than a feature of the system.)

### Visual Analytics for Multimodal Social Network Analysis: A Design Study with Social Scientists

Primary Reviewer: Bruno

Secondary Reviewer: Fintan

#### Secondary Review

This is a very important design study paper for the project, and was a key paper in inspiring the project., The term multimodal was our initial descriptor for the topic but has been superseded in our discussion by multilayer. The layers are defined by node type. The authors also relate the concept to n-partite graphs ( especially in terms of analysis) .The authors refer to a node link view is standard but they also have a novel view called parallel node link bands, which are essentially Parallel coordinates where nodes are stored and replicated across axes and links are only drawn in adjacent bands. The ego networks of nodes (Within mode) connections can be shown on demand. The PNLB provides some novel interaction e.g. sorting nodes based on attributes or connectivity to other bands, show ego network on demand, usual hovering functionality etc . They also perform some novel attribute visualization, e.g. Wordclouds of text associated with nodes, multivariate display of data using an embedded mini parallel coordinates in the PNLB.

As part of their user study the also identify some interesting tasks, e.g. Identify extreme nodes Correlate within and between layer activities, Correlate attributes and between layer links, Correlate attributes and within layer links. For analytics the provide Multimodal specific (I.e. n- partite) versions of common centralities (e.g. degree , closeness , betweenness). The only negative point is the on fact they do not include edge type as part of their analysis.

### TopoLayout: Multilevel Graph Layout by Topological Features

Primary Reviewer: Fintan The focus of this paper is identifying topological features and laying them out as subgraphs , with the most appropriate algorithm.

This paper does not really related to ML stuff and is not really needed in our surveyr it was only included as we cited it in the proposal, but that citation did not concern an ML functionality. It might be interesting if we were discussing different layout approaches related to structure in ML graph, but that is it

Secondary Reviewer: Bruno

### behaviorism: a framework for dynamic data visualization

Primary Reviewer: Sebastien

Secondary Reviewer: Bruno

### FluxFlow: Visual Analysis of Anomalous Information Spreading on Social Media

Secondary Reviewer: Guy

### Visualization of Heterogeneous Data

Primary Reviewer: Fintan

Secondary Reviewer: Sebastein

#### Primary Review:

This paper is not about visualization and the data is not heterogeneous (in the multilayer sense). Heterogeneous refers many different types of data , with attributes missing. NOT really layers. Its reall about using RDF queries to match poor quality data to an appropriate technique The visualizations shown are not a part of the contribution. It should not be included in our survey.

### Visual Analysis of Higher-Order Conjunctive Relationships in Multidimensional Data Using a Hypergraph Query System

Primary Reviewer: Bruno

Secondary Reviewer: Fintan

### The FlowVizMenu and Parallel Scatterplot Matrix: Hybrid Multidimensional Visualizations for Network Exploration

Primary Reviewer: Sebastien

Secondary Reviewer: Fintan

#### Primary Review:

The technique combines multiple approaches, scatter plots, parallel coordinates and Parallel ScatterPlot Matrices, allowing to convert from a scatter plot representation to a parallel coordinate representation. A novel interaction (popup widget) allows to fluidly select node attributes and also generate on-the-fly proper scatter plots. Scatter plots may be used for domain specific attributes (protein name, biological function, cellular localization) or generic network attributes (degree, betweenness, clustering coefficients). This was applied for generic network exploration but could be used to compare layers attributes as well. Evaluation is limited but interaction is interesting.

#### Secondary Review:

This paper is a mixed bag of techniques without a strong specific focus. It features novel visualizations derived from scatterplot matrices and parallel coordinates and is primarily related to node attributes, however they could be adapted to include attributes across layers. However these are minor tweaks and not ground breaking, but they could also be adapted for multilayer attribute visualization, where we want to compare attributes across layers, so are worth mentioning in our survey. There is very little empirical evaluation. The interaction technique is interesting, but not very multilayer focused.

### Flow Mapping and Multivariate Visualization of Large Spatial Interaction Data

Primary Reviewer: Guy

Secondary Reviewer: Bruno

#### Primary Review:

@article{RN2079,

 author = {Guo, Diansheng},
title = {Flow mapping and multivariate visualization of large spatial interaction data},
journal = {IEEE Transactions on Visualization and Computer Graphics},
volume = {15},
number = {6},
ISSN = {1077-2626},
year = {2009}

}

The paper presents algorithms, visual encodings and interactions on maps depicting flows between geographically based entities. Although we could consider the data as being a multilayer network, it is not treated as such, but more as a multivariate network (and we indeed need to explain how these two things differ). Entities are clustered based on a given taxonomy (regions), flows are aggregated using a SOM approach. Parallel coords is used to give insight on how flows were aggregated (computing “regions”).

### Interactive Exploration of Implicit and Explicit Relations in Faceted Datasets

Secondary Reviewer: Sebastien

### GraphCharter: Combining browsing with query to explore large semantic graphs

Primary Reviewer: Bruno

#### Secondary Review

This paper presents a system to query and browse large semantic graphs e.g Freebase or wikipedia. These graphs hold hundred of node and edge types, and are in fact multilayer graphs even though the authors don't consider any notion of layer per se. The system provides a visual interface where the analyst can add meta-nodes/edges and instance nodes/edges pulled from a database. Light-weight queries allow the analyst to build a graph of interest incrementally following a “query, expand, query again” method. This approach is seen as complementary to the use of a DOI expansion mechanism insofar that the expansion step can leverage DOI functions to rank its results. The query building is carried out interactively on top of a simple node-link visualization of the graph at hand using contextual menus. Meta-elements are decorated with numbers hinting to the number of instance nodes/edges that could potentially be returned upon expansion. The graph browsing interactions under consideration are:

• showing node properties,
• expanding a meta edge from an instance/meta node,
• expanding a meta node by adding corresponding instance nodes; to reduce the interaction load, an auto expansion heuristic is used, which favors high degree nodes. This can however be replaced by a proper DOI algorithm.
• copy/paste of nodes and their associated meta-edges
• standard pan/zoom, single and multiple selection of nodes/edges and drag and drop,

The system uses standard FR layout but locks the position of previously existing nodes upon expansion. A layered layout (in the sense of Battista et al. book) is also offered, resulting in a tree-like representation. The visual encoding consists in using color for element type, and using different shapes/labels to distinguish nodes from meta-nodes and line thickness and patterns to distinguish edges from meta-edges.

The paper presents a case study and user study to assess the usefulness of the system.

This work provides an interesting list of interactions for graph specification and incremental browsing, which may nicely complement the interactions described by Hascoet and Dragicevic (about interactive graph matching). Even if the authors don't take a multilayer graph perspective, and don't materialize any visual layers, we can draw a parallel with some of our constructs and interactions.

### Interactive high-dimensional visualization of social graphs

Secondary Reviewer: Guy

### MultiStory: Visual analytics of dynamic multi-relational networks

Primary Reviewer: Sebastien

Secondary Reviewer: Benoit

#### Primary Review:

The paper presents a coordinated view between a “story line”, and a node-link view. Some (traditional) interactions techniques are presented. There are two small studies used to illustrate the tool, in particular to show it's useful in terms of visual analytics. This allow to visualize for instance phone communication patterns of friendship clusters, at some points in time. The approach have some limitations ; it's difficult to analyze many correspondences at once, it only uses classical techniques, and has no formal evaluation. The paper could serve to provide examples of use cases and tasks as part of the taxonomy to build (in this case. to visually analyze multilayer social networks). The presented technique itself does not appear sufficient but do cover evolution of multiple types of edges (multilayer aspect).

### Multilayer graph edge bundling

Primary Reviewer: Guy

Secondary Reviewer: Fintan

#### Secondary Review

This is an interesting application of an existing technique (edge Bundling) to a multilayer ( or at least multi relational) use case. The presentation is a standard 2d Layout, There is only a single graph vis but the different types of edges are bundled together within it in a readable way. The data sets are Social Networks (relationship types defined by interactions eg (call, text, friendship etc…) and BIOGRID, a protein interaction network where the relationships were not specified. The layer definition is always edge type. The main focus is on edge routing, with no exciting interaction or tasks, or any real evaluation beyond images of results (although the technique does look effective) Definitely worth citing as an example for an existing technique applied to a multilayer graph use case.

==== Visual exploration of Location-Based Social Networks data in urban planning ====

Primary Reviewer: Bruno

#### Secondary Review

Irrelevant, not a graph visualization paper.

### GraphPrism: Compact Visualization of Network Structure

Primary Reviewer: Sebastien

Secondary Reviewer: Guy

### Focus Dependent Multi-level Graph Clustering

Primary Reviewer: Guy

Secondary Reviewer: Bruno

### Interactive Graph Matching and Visual Comparison of Graphs and Clustered Graphs

Secondary Reviewer: Guy

#### Primary Review:

This paper presents an integrated system for graph matching and comparison based on a multi-layer interaction model. A layer is defined as a translucent 2D visual plane augmented with a graph structure and its rendering properties (a layout, an animation, geometric transformations). This leads to a unification of the main previous visual representation approaches, namely small multiples, stacked/merged layers and animation. An important contribution of this paper is to offer means to manipulate and tune layers both individually and group-wise. The system supports reconfigure and explore interaction techniques (as defined by Yi et al.). Concretely, reconfigure interactions consist within one layer in rearranging node positions based on a variety of layout options (optimization-based, attribute-based) including the possibility to apply several layouts in sequence (transformation-based) in order to improve the final layout and the possibility to introduce layout perturbations (a hole, projection on a shape). Constraining the position of nodes in one layer based on their position in another layer is also supported, and aims to facilitate graph comparison. In complement, exploration techniques consist in zoom, pan, and animations. Within one layer, zoom can be applied separately to node size and node distance (spreading nodes apart). Zoom level can be adjusted at the layer level or for a group of layers. Layers can be put side-by-side or stacked, and moved across windows using drag and drop. Within one layer, layout algorithms are animated progressively and can be interrupted or triggered in sequence based on user interaction. Flip-book animations allow browsing through multiple layers using a crossing-based interaction, in order to support graph comparison. The reificiation of graph matching functions allows the creation of a master graph where nodes are grouped across layers based on their proximity in the 2D space of their respective layers. This master graph can then be manipulated in a separate layer and used to constrain the other layers. Manual node matching allows a more semantic/expertise-driven editing of the master graph. The usefulness of the system is validated with 3 use cases. In the first, the temporal evolution of topics in the infovis community is discussed using a graph of keywords. A layer is created for distinct years. In the second use case, the system is used to inspect a large lexical network containing hyponymy-hypernymy relationships generated using crowdsourcing. The layers are created based on either the type of relationship, or timestamp, depending on the type of analysis being carried out. Insights can be obtained about missing links, direction of growth. In the third use case, the system is used to compare the results of different graph clustering algorithms. Using small multiples, it is possible to contrast overall cluster structure and cluster size homogeneity. Stacking the layers allows a more detailed comparison of cluster membership.

• high-level tasks: highlight the matches and differences between graphs;
• means of comparison: 2D stacking of layers with layer-specific settings, small multiples, animations;
• layer interactions: show/hide layers, modify the order of layers, zoom and pan layers individually or together

Layer specification/creation is not described clearly. Not sure whether it is done in the tool or using an external processing tool.

### Wakame: Sense Making of Multi-dimensional Spatial-temporal Data

Primary Reviewer: Sebastien

Secondary Reviewer: Guy

### Social-Circles Exploration Through Interactive Multi-Layered Chord Layout

Primary Reviewer: Bruno

#### Secondary Review

This short paper proposes a multi-level chord layout representation aiming to support the exploration of ego-networks and at the same time the social circles to which the nodes belong. It combines traditional chord layout with a hierarchy built on categorical attributes of the nodes (the so-called social circles), like multi-level geographic information Country→Region→City or music→genre→band. Nodes may repeat across layers/social circles. Dynamic labeling shows details on demand. Selection and filtering are used to de-emphasize the parts of the network which is not connected to the selected node(s). Selecting a circle de-emphasizes the nodes not belonging to it, except optionally the immediate neighbors. This is an early stage work, lacking any evaluation.

### g-Miner: Interactive Visual Group Mining on Multivariate Graphs

Primary Reviewer: Bruno

#### Secondary Review

This paper presents a system to visualize and refine the results of a group mining algorithm from within multivariate graphs. It builds a hierarchy of clusters visualized as an icicle plot to explore the graph. Focus groups can be visualized as NL diagrams to explore internal relationship in a matrix view and attribute values can be seen as a heatmap. The graphs at hand are not multilayer. But an analogy between a group/cluster and a layer might be drawn, which is somewhat the definition of layer used in the original hive plots paper. The usefulness of the system is evaluated through a case study and interviews with domain experts.

From a task perspective, group creation can be carried out by 1) using the group hierarchy, 2) building an ego network for a node of interest found using search, 3) template matching based on members attributes, connection or group structure. Group editing/refinement by changing the group creation criteria.

### The Effects of Representation and Juxtaposition on Graphical Perception of Matrix Visualization

Primary Reviewer: Sebastien

Secondary Reviewer: Guy

### GraphTrail: analyzing large multivariate, heterogeneous networks while supporting exploration history

Primary Reviewer: Benoit

### A multilevel analysis of sociability, usability, and community dynamics in an online health community

Secondary Reviewer: Benoit

### Visual Analysis of Trajectories in Multi-Dimensional State Spaces

Primary Reviewer: Fintan

Secondary Reviewer: Sebastien

#### Primary Review:

This paper focuses on attribute visualization across several dimensions, usually multiple time slices. It is not a network visualization paper. However the thee techniques may be of some interest if we are looking at time based data over layers, or looking at comparing layers The first is a continuous time scatter plot. It is a scatterplot where rather than sampling values at certain timepoints, the data is integrated to get trajectories and lines are drawn on the plot. The second is continuous time parallel coordinates, that simply use a different approach to calculating the density to reflect the continuous nature of the data. The most interesting is the Temporal heatmap, which is essentially a heat map where every column is a variable, the vertical axes represent time, and colour represents variable value. This could be a very interesting way of comparing layers. Overall an interesting paper on visualizing multidimensional attribute data where there is trajectory information ( i.e. a continuous underlying data model)

As evaluation they offer 3 case studies, without any empirical evaluation. The visual approach described by the temporal heat map may offer an interesting way of comparing attribute values across layers.

### Detangler: Visual Analytics for Multiplex Networks

Secondary Reviewer: Fintan

#### Secondary Review

This is a very important paper important for interaction, Layer definition and attribute vis. It focuses on very interesting interaction, which is relevant to cross layer interactions. There is very nice attribute visualization combined with selection (simple but effective) and very important approach to layer visualization. The data sets are a Historical Social Network, and a document data set realted by common terms. In their explanatory example the topics of papers are used to create multiplex links. Catalyst and substrate networks are also created, and these can be considered a form of layer. Catalyst networks are networks where co-occurring topics are represented. The authors also provide usage scenarios with road safety and historical data sets. A separate layer is made based on relationship type ( the substrate layer)“ Multiple Edge types characterise the multiplex graphs. A separate Catalyst layer is created along with the original graph, that shows how the edge types relate to each other A node view is used for visualization. There is a linked view of the original ( catalyst graph) and the substrate. The layouts between the two are harmonised As a type of interactive attribute visualization there is a selection lasso which is coloured based on the ”“entanglement”“ score for a set of nodes. The leapfrog interaction is very interesting as well, all owing a selection to be made based on relationships in the other layer. As an interesting form of attribute vis the selection lasso is coloured based on the “entanglement” score for a set of nodes. This is a nice mix of interaction and attribute vis. Also attribute size is adjusted based on certain scores, and standard highlighting. Finding relationships relative to a node set, and finding node sets relevant to a relationship set ( via Leapfrogging) The entanglement score while based on some of the authors previous work is novel There is some evaluation as detailed usage scenarios are provided, as well as a description of an expert feedback session. No perceptual or standard task based time an error evaluation is done

### AVOCADO: Visualization of Workflow–Derived Data Provenance for Reproducible Biomedical Research

Primary Reviewer: Guy

Secondary Reviewer: Bruno

### Topology-based Visualization of Transformation Pathways in Complex Chemical Systems

Primary Reviewer: Bruno

Secondary Reviewer: Fintan

### GraphDice: A System for Exploring Multivariate Social Networks

Primary Reviewer: Sebastien

### Refinery: Visual Exploration of Large, Heterogeneous Networks through Associative Browsing

Primary Reviewer: Fintan

Secondary Reviewer: Bruno

#### Primary Review:

This is a good system well implemented, one of the primary contribution is novel use of DOI to explore heterogeneous data. Over this paper is important to our START and BLIZAAR as a whole The data is they discuss in theory truly heterogeneous as in the type is textual , or non textual such as multimedia. etc. However in practice, with their example data set they define nodes based on a type field. The authors present a combined node-link and list approach. This is not exceptionally novel , but the way the list and Nodleink interact, especially with the search exploration functionality, is effective. The structure of the list is also a nice approach ( group list by facets) For exploring the data the authors represent an interesting upvote/downvote system that influences their random walk approach, in addition to keyword querying. Their approach for finding new data is a novel random walk DOI algorithm The grouping of the list by type / facet on the list view is interesting. Beyond that they simply use a pretty NL layout with node size and standard highlights. The main application is exploration of data across multiple types (or layers) and their approach to this task, a random walk based doi capable of handling heterogeneous data. In terms on analytic contributions, the random walk DOI can be considered an analytic approach. They also provide a decent evaluation. It is a user study featuring academic and industry researchers, resulting in behavioural observations and feedback. Results reflect actual system usage and subjective feedback. One minor negative is that the heterogeneity of the data isn't as much of a focus as it should be. This leaves a lot of scope for focusing on multilayer issues in the BLIZAAR project

### Visual Analysis of Governing Topological Structures in Excitable Network Dynamics

Primary Reviewer: Guy

### Visualizing Multidimensional Data with Glyph SPLOMs

Primary Reviewer: Benoit

Secondary Reviewer: Sebastien

### Heterogeneous Networks on Multiple Levels

Primary Reviewer: Fintan

Secondary Reviewer: Guy

### Multivariate Networks in the Life Sciences

Primary Reviewer: Guy

#### Secondary Review

This paper introduces the need of network visualization in life sciences. It starts by characterizing -omics data and related tasks. The focus of analysts may be on a single network, the interconnection between different networks or to compare multiple networks at the same time. They may also associate a variety of data with either the nodes or the links, hence the multivariate nature. The phenomena under investigation range from the atom level up to the species level, warranting the use of hierarchical/multilevel navigation. Some entities may repeat in different network types or layers, which seem to be defined based on edge type. The paper also mentions the availability of multiple biological databases, which are enriched gradually. Yet, analysts are often more interested by changes in the sate of the network rather than structural changes. Authors also point out that only parts of the network are relevant at a given time e.g. parts featuring statistically significant changes in expression levels. The paper discusses a few use cases (signaling pathways, genetic linkage, document-based relationship discovery and gene regulatory networks). The visualizations envisioned in the paper include:

• a schema in the form of a hierarchical node-link diagram, with homogeneous nodes at each level, and color being used to encode node type; links are shown within each layer and between layers.
• small multiples of node-link diagrams for each omic layer. Links between layers point out shared nodes, which is redundantly encoded using color.
• in presence of timeseries data, a linechart diagram can be embedded in the related nodes of a node-link representation
• heatmap matrices are propular ways to visualize expression/correlation data

The paper ends with a list of challenges related to biological data:

• scalability;
• uncertainty of the structure but also the related attributes;
• heterogeneity: different types od nodes, edges, hyper edges, and hierarchical relationships;
• interactivity is key to large-scale data exploration;
• standardization, as in standardized glyphs for different node and link types, e.g. using the SBGN notation.

### Temporal Multivariate Networks

Primary Reviewer: Bruno

### Multilevel visualization of clustered graphs

Secondary Reviewer: Sebastien

### Orion: A system for modeling, transformation and visualization of multidimensional heterogeneous networks

Primary Reviewer: Guy

### Ploceus: Modeling, visualizing, and analyzing tabular data as networks

Primary Reviewer: Bruno

Secondary Reviewer: Guy

### A Novel Visualization Technique for Electric Power Grid Analytics

Primary Reviewer: Benoit

Secondary Reviewer: Guy

### Multivariate Network Exploration with JauntyNets

Primary Reviewer: Sébastien

Secondary Reviewer: FIntan

#### Primary Review:

The tool developed in this research (application-based paper) includes an attribute-driven layout to support multiple (node) attributes contrary to existing force-based layouts. They combined multiple coordinated views with attribute-driven layout. Using the approach, node position is related to the corresponding attributes values. Node attributes are shown using bar charts Clustering and multidimensional scaling is also supported. The authors focus on node attributes, edge attributes are mostly ignored (but edge can be weighted). Document (nodes) correlating with a group of attributes move towards that group.

### The structure and dynamics of multilayer networks

Primary Reviewer: Guy

Secondary Reviewer: Bruno

### System-of-systems formulation and disruption analysis for multi-scale critical national infrastructures

Secondary Reviewer: Sebastien

### Discovering suspicious behavior in multilayer social networks

Primary Reviewer: Fintan

### Towards effective visual analytics on multiplex and multilayer networks

FYI this looks like it could be the paper the book chapter above is based on.

Secondary Reviewer: Guy

### TMNVis: Visual Analysis of Evolution in Temporal Multivariate Network at Multiple Granularities

Primary Reviewer: NA

Secondary Reviewer: NA

Removed from survey as was found using multivariate graphs as a query

### Multilayer dynamics of complex spatial networks: The case of global maritime flows (1977–2008)

Primary Reviewer: Bruno

Secondary Reviewer: Fintan

#### Secondary Review

This paper primarily concerns presenting a data set concerning shipping traffic. It is a multiplex geographic network with different freight types represented by different edge types. Analytics is mostly within the layer, but he does apply some multiplex analytics. e.g. they use an edge based variant of homiphily( how many types of node of the same type are connected, except the types of the edges routed through the node. The goal is to identify hubs. He uses a standard of he shelf visualization to produce a nodelink visualization with multiple layer types. His example visualization illustrate clearly how the application domains could benefit frrm better vis

### Network-Thinking: Graphs to Analyze Microbial Complexity and Evolution

Primary Reviewer: Fintan

### Hierarchical Focus+Context Heterogeneous Network Visualization

Primary Reviewer: Fintan

Secondary Reviewer: Bruno

### MultiNets: Web-Based Multilayer Network Visualization

Secondary Reviewer: Bruno

### NodeTrix-Multiplex: Visual Analytics of Multiplex Small World Networks

Primary Reviewer: Bruno

Secondary Reviewer: Fintan

### The Multiplex Network of EU Lobby Organizations

Primary Reviewer: Fintan

Secondary Reviewer: Sebastien

#### Primary Review:

This is data set paper form the domain of complex networks. It consists primarily of presenting a dataset and associated analytics. It uses a standard node link visualization, and it is shown as a network overview. The paper is more focused on analaytics, and for the attribute visualization The authors use a graduated rainbow scale for categorisation of attribute type in their attribute scatterplots, which perhaps might server as an example of how standard practices in the visualization community have not percolated through to application domains. Its primary contribution is in the analysis of the data and how the authors define the layers. The data has been gathered from publically available sources and it is a multiplex dataset where nodes represent lobbying organisations. While organisations are categorised by domain, this is not used to define layers.

The author examine each layer individually for patterns. They also use a centrality metric that is calculated across all layers. It is a custom metric ( based on PageRank) with the goal to determine the influence of the lobbying organisations

### Hive plots—rational approach to visualizing network

Primary Reviewer: Guy

Secondary Reviewer: FIntan

### Visual Reasoning about Social Networks Using Centrality Sensitivity

Secondary Reviewer: Guy

### The multilayer temporal network of public transport in Great Britain

Primary Reviewer: Bruno

Secondary Reviewer: Fintan

#### Secondary Review

This is primarily a dataset paper, and it uses muxvis to visualize its data. Its main purpose is to provide a data set, but the way it builds that data set is very relevant. It contains 6 layers (Rail, Air, Coach, Ferry, Metro, bus). The data also has a temporal aspect. So it contains no novel interaction, or vis etc. However, it is an interesting example of a publically available multilayer Network dataset, so is relevant to our survey

### Clustering With Multi-Layer Graphs: A Spectral Perspective

Primary Reviewer: Fintan

Secondary Reviewer: Guy

#### Secondary Review

Primary Reviewer: Benoit

Secondary Reviewer: Sebastien

### Graphiti: Interactive Specification of Attribute-Based Edges for Network Modeling and Visualization

Primary Reviewer: Fintan

Secondary Reviewer: Guy

### Vislink: Revealing relationships amongst visualizations.

Primary Reviewer:

Secondary Reviewer:

### Example Paper Title

Primary Reviewer:

Secondary Reviewer: